TalkToMe

Talk to a person based on a Transformer model of them.

This application scrapes chat logs between two users from Discord and inputs one of the user's data through a GPT-2-medium model to generate a relatively complex language model. The dataset, found under training_2_.txt, contains 300 lines of conversation between Mykola and Steven, two friends from UCF. Because the training data is multilingual, GPT-2 can still adapt to these changes and generate fairly reasonable text. For training, we use the gpt-2-simple library and fine tune the model to our own data.

Web Application Used

We utilize Python's Flask as our backend to receive user's input text. It passes it through the GPT-2 model to retrieve the output and forwards to the front-end UI. The front-end was inspired by this project and we use his front-end, developed with HTML, CSS and JavaScript as our template.

Link to Model:

The model that we trained can be found here

Running Web Application

1. cd web_app
2. pip install -r requirements.txt
3. python app.py
4. Type localhost:5000 in your favorite browser and you're ready to go! Express yourself to yourself!